Extraction of Motion Artifacts from the Measurements of a Wireless Electrocardiogram using Tensor Decomposition

Wireless electrocardiograms facilitate long-term and affordable monitoring of patients in their residential and work environments. However, the measurements are often affected by motion artifacts whose statistical properties are difficult to model or estimate. In this paper we employ tensor decomposition to model motion artifacts. Our approach is based on two steps. Firstly, we transform the ECG measurements into time-frequency space using the continuous wavelet transform and construct a three-dimensional array (the wireless ECG provides three different channels and the wavelet transform of each channel generates a two-dimensional matrix encoding spectral and temporal aspects). Secondly, we employ tensor decomposition to factorize the data and extract motion patterns. We consider different types of movements associated with everyday activities and validate our method by comparing the correlation between the motion artifacts we establish with inertial measurements which are taken in sync with the ECG measurements.